tfjs-v3.20.0

What's Changed

• [Jasmine] Add expectation to tests in jasmine test framework by @haoyunfeix in #6656
• Set willReadFrequently to true for GPU correctness test by @gyagp in #6665
• [WebGL] Implement packed NHWC Conv2d by @Linchenn in #6639
• [WebGL] Fix NHWC packed depthwise conv2d for dilation=3 by @Linchenn in #6662
• webgpu: dont use mod as it is a reserved keyword by @crowlKats in #6667
• [converter] updated the dev readme for converter to cover python tests by @pyu10055 in #6675
• [core] Fix undefined expect in test_util by @haoyunfeix in #6671
• [WebGL] Fix the test for Depthwise with dilation=3 by @Linchenn in #6679
• webgpu: fix notEqual error by @xhcao in #6669
• [local-benchmark tool] Enable populate modelUrl and input shapes via URL by @Linchenn in #6680
• Add iOS and Android to WebGL presubmit tests by @mattsoulanille in #6681
• [webgpu] Remove extra device stuff passing to the constructor by @haoyunfeix in #6689
• Fix browserstack-benchmark tool by @Linchenn in #6703
• [webgpu] Set WEBGPU_IMPORT_EXTERNAL_TEXTURE to true by default by @axinging in #6706
• [webgpu] Fix alignment missing for 5D and 6D by @haoyunfeix in #6701
• [browserstack benchmark tool] Enable customizing local builds by @Linchenn in #6705
• Use the firebase-admin package instead of firebase to push benchmark results by @mattsoulanille in #6711
• [browserstack benchmark tool] Enable code snippet benchmark by @Linchenn in #6704
• [browserstack benchmark tool] Update browser list with new devices by @Linchenn in #6713
• Make link-package's build-deps-for script check transitive deps by @mattsoulanille in #6712
• Add structured outputs support to tfjs-inference by @ahmedsabie in #6696
• [browserstack benchmark tool] fix build script by @Linchenn in #6716
• Run the next BrowserStack benchmark immediately instead of in batches by @mattsoulanille in #6717
• Add kernel RaggedTensorToTensor for CPU and WebGL backend by @ahmedsabie in #6686
• Fix tf.image.transform on CPU and WASM by @andrusza2 in #6658
• [automl] Use the same karma timeout settings as Bazel by @mattsoulanille in #6719
• De-flake fromPixels test failure by @hujiajie in #6626
• [tfjs-converter] Revert "fix protobuf version to 3.20.0 (#6612)" by @pyu10055 in #6721
• [browserstack benchmark tool] End firebase instance only if it exists by @Linchenn in #6727
• Setup optimization target deices/browsers by @Linchenn in #6722
• Remove fromPixelsAsync test cases by @gyagp in #6723
• webgpu: Skip raggedTensorToTensor test cases by @gyagp in #6732
• webgpu: export webgpu utilities by @xhcao in #6707
• [benchmark tools] Update mobilenet models by @Linchenn in #6729
• [benchmark tools] Update model lists for benchmarking by @Linchenn in #6730
• [tfjs-converter] fixed protobuf version mismatch error on cloudbuild by @pyu10055 in #6737
• [local benchmark tool] Add convertedBy filed by @Linchenn in #6745
• Add use_structured_outputs flag to README by @ahmedsabie in #6739
• [browserstack benchmark tool] Add a html to present benchmark results by @Linchenn in #6740
• [browserstack benchmark tool] Move CodeSnippet from backend to frontend by @Linchenn in #6738
• webgpu: Merge MatMulPackedProgram and MatMulPackedVec4Program by @qjia7 in #6688
• [browserstack benchmark tool] batch benchmark code snippet by @Linchenn in #6742
• [webgpu] Add Atan2, IsNaN, Reciprocal by @axinging in #6743
• [tfjs-converter] remove kept flag after tensor is replaced or popped from the tensor list by @pyu10055 in #6759
• Strengthen frame readiness while testing fromPixels with video element by @hujiajie in #6751
• webgpu: enlarge the splitted dimInner size by @qjia7 in #6755
• [webgpu] s/ScatterOptimizedProgram/ScatterProgram/g by @hujiajie in #6761
• Adds JAX-->TFjs converter by @marcvanzee in #6744
• [browserstack benchmark tool] Use stream to write to outfile by @Linchenn in #6750
• [tfjs-converter ]include jax_conversion target to fix nightly test by @pyu10055 in #6767
• [WebGL backend] Add a flag to enable Software WebGL by @Linchenn in #6772
• [webgpu] Tweak WGSL entry point emission so '{' isn't visually swallowed by @hujiajie in #6764
• Add cpu forward for Cast by @qjia7 in #6760
• webgpu: Optimize AvgPool when filter size = input size by @qjia7 in #6762
• [op] allow dtype secification for onehot op by @pyu10055 in #6781
• webgpu: Add non-shared argminmax program by @qjia7 in #6778
• [converter] added dtype support for oneHot for converter by @pyu10055 in #6782
• Update monorepo to 3.20.0. by @pyu10055 in #6788
• Update lockfiles branch tfjs_3.20.0_lockfiles lock files. by @pyu10055 in #6789

New Contributors

• @andrusza2 made their first contribution in #6658
• @marcvanzee made their first contribution in #6744

Full Changelog: tfjs-v3.19.0...tfjs-v3.20.0